redundancy-elimination.cc

Go to the documentation of this file.

// Copyright 2016 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
 
#include "src/compiler/redundancy-elimination.h"
 
#include "src/compiler/js-graph.h"
#include "src/compiler/node-properties.h"
#include "src/compiler/simplified-operator.h"
 
namespace v8 {
namespace internal {
namespace compiler {
 
RedundancyElimination::RedundancyElimination(Editor* editor, JSGraph* jsgraph,
                                             Zone* zone)
    : AdvancedReducer(editor),
      node_checks_(zone),
      jsgraph_(jsgraph),
      zone_(zone) {}
 
RedundancyElimination::~RedundancyElimination() = default;
 
Reduction RedundancyElimination::Reduce(Node* node) {
  if (node_checks_.Get(node)) return NoChange();
  switch (node->opcode()) {
    case IrOpcode::kCheckBigInt:
    case IrOpcode::kCheckedBigIntToBigInt64:
    case IrOpcode::kCheckBounds:
    case IrOpcode::kCheckClosure:
    case IrOpcode::kCheckEqualsInternalizedString:
    case IrOpcode::kCheckEqualsSymbol:
    case IrOpcode::kCheckFloat64Hole:
    case IrOpcode::kCheckHeapObject:
    case IrOpcode::kCheckIf:
    case IrOpcode::kCheckInternalizedString:
    case IrOpcode::kCheckNotTaggedHole:
    case IrOpcode::kCheckNumber:
    case IrOpcode::kCheckNumberFitsInt32:
    case IrOpcode::kCheckReceiver:
    case IrOpcode::kCheckReceiverOrNullOrUndefined:
    case IrOpcode::kCheckSmi:
    case IrOpcode::kCheckString:
    case IrOpcode::kCheckStringOrStringWrapper:
    case IrOpcode::kCheckSymbol:
    // These are not really check nodes, but behave the same in that they can be
    // folded together if repeated with identical inputs.
    case IrOpcode::kStringCharCodeAt:
    case IrOpcode::kStringCodePointAt:
    case IrOpcode::kStringFromCodePointAt:
    case IrOpcode::kStringSubstring:
#define SIMPLIFIED_OP(Opcode) case IrOpcode::k##Opcode:
      SIMPLIFIED_CHECKED_OP_LIST(SIMPLIFIED_OP)
      SIMPLIFIED_BIGINT_BINOP_LIST(SIMPLIFIED_OP)
#undef SIMPLIFIED_OP
      return ReduceCheckNode(node);
    case IrOpcode::kSpeculativeNumberEqual:
    case IrOpcode::kSpeculativeNumberLessThan:
    case IrOpcode::kSpeculativeNumberLessThanOrEqual:
      return ReduceSpeculativeNumberComparison(node);
    case IrOpcode::kSpeculativeNumberAdd:
    case IrOpcode::kSpeculativeNumberSubtract:
    case IrOpcode::kSpeculativeAdditiveSafeIntegerAdd:
    case IrOpcode::kSpeculativeAdditiveSafeIntegerSubtract:
    case IrOpcode::kSpeculativeSmallIntegerAdd:
    case IrOpcode::kSpeculativeSmallIntegerSubtract:
    case IrOpcode::kSpeculativeToNumber:
      return ReduceSpeculativeNumberOperation(node);
    case IrOpcode::kEffectPhi:
      return ReduceEffectPhi(node);
    case IrOpcode::kDead:
      break;
    case IrOpcode::kStart:
      return ReduceStart(node);
    default:
      return ReduceOtherNode(node);
  }
  return NoChange();
}
 
// static
RedundancyElimination::EffectPathChecks*
RedundancyElimination::EffectPathChecks::Copy(Zone* zone,
                                              EffectPathChecks const* checks) {
  return zone->New<EffectPathChecks>(*checks);
}
 
// static
RedundancyElimination::EffectPathChecks const*
RedundancyElimination::EffectPathChecks::Empty(Zone* zone) {
  return zone->New<EffectPathChecks>(nullptr, 0);
}
 
bool RedundancyElimination::EffectPathChecks::Equals(
    EffectPathChecks const* that) const {
  if (this->size_ != that->size_) return false;
  Check* this_head = this->head_;
  Check* that_head = that->head_;
  while (this_head != that_head) {
    if (this_head->node != that_head->node) return false;
    this_head = this_head->next;
    that_head = that_head->next;
  }
  return true;
}
 
void RedundancyElimination::EffectPathChecks::Merge(
    EffectPathChecks const* that) {
  // Change the current check list to a longest common tail of this check
  // list and the other list.
 
  // First, we throw away the prefix of the longer list, so that
  // we have lists of the same length.
  Check* that_head = that->head_;
  size_t that_size = that->size_;
  while (that_size > size_) {
    that_head = that_head->next;
    that_size--;
  }
  while (size_ > that_size) {
    head_ = head_->next;
    size_--;
  }
 
  // Then we go through both lists in lock-step until we find
  // the common tail.
  while (head_ != that_head) {
    DCHECK_LT(0u, size_);
    DCHECK_NOT_NULL(head_);
    size_--;
    head_ = head_->next;
    that_head = that_head->next;
  }
}
 
RedundancyElimination::EffectPathChecks const*
RedundancyElimination::EffectPathChecks::AddCheck(Zone* zone,
                                                  Node* node) const {
  Check* head = zone->New<Check>(node, head_);
  return zone->New<EffectPathChecks>(head, size_ + 1);
}
 
namespace {
 
struct Subsumption {
  enum class Kind {
    kNone,
    kImplicit,
    kWithConversion,
  };
 
  static Subsumption None() { return Subsumption(Kind::kNone, nullptr); }
  static Subsumption Implicit() {
    return Subsumption(Kind::kImplicit, nullptr);
  }
  static Subsumption WithConversion(const Operator* conversion_op) {
    return Subsumption(Kind::kWithConversion, conversion_op);
  }
 
  bool IsNone() const { return kind_ == Kind::kNone; }
  bool IsImplicit() const { return kind_ == Kind::kImplicit; }
  bool IsWithConversion() const { return kind_ == Kind::kWithConversion; }
  const Operator* conversion_operator() const {
    DCHECK(IsWithConversion());
    return conversion_op_;
  }
 
 private:
  Subsumption(Kind kind, const Operator* conversion_op)
      : kind_(kind), conversion_op_(conversion_op) {
    DCHECK_EQ(kind_ == Kind::kWithConversion, conversion_op_ != nullptr);
  }
 
  Kind kind_;
  const Operator* conversion_op_;
};
 
// Does check {a} subsume check {b}?
Subsumption CheckSubsumes(Node const* a, Node const* b,
                          MachineOperatorBuilder* machine) {
  Subsumption subsumption = Subsumption::Implicit();
  if (a->op() != b->op()) {
    if (a->opcode() == IrOpcode::kCheckInternalizedString &&
        b->opcode() == IrOpcode::kCheckString) {
      // CheckInternalizedString(node) implies CheckString(node)
    } else if (a->opcode() == IrOpcode::kCheckString &&
               b->opcode() == IrOpcode::kCheckStringOrStringWrapper) {
      // CheckString(node) implies CheckStringOrStringWrapper(node)
    } else if (a->opcode() == IrOpcode::kCheckInternalizedString &&
               b->opcode() == IrOpcode::kCheckStringOrStringWrapper) {
      // CheckInteralizedString(node) implies CheckStringOrStringWrapper(node)
    } else if (a->opcode() == IrOpcode::kCheckSmi &&
               b->opcode() == IrOpcode::kCheckNumber) {
      // CheckSmi(node) implies CheckNumber(node)
    } else if (a->opcode() == IrOpcode::kCheckSmi &&
               b->opcode() == IrOpcode::kCheckNumberFitsInt32) {
      // CheckSmi(node) implies CheckNumberFitsInt32(node)
    } else if (a->opcode() == IrOpcode::kCheckNumberFitsInt32 &&
               b->opcode() == IrOpcode::kCheckNumber) {
      // CheckNumberFitsInt32(node) implies CheckNumber(node)
    } else if (a->opcode() == IrOpcode::kCheckedTaggedSignedToInt32 &&
               b->opcode() == IrOpcode::kCheckedTaggedToInt32) {
      // CheckedTaggedSignedToInt32(node) implies CheckedTaggedToInt32(node)
    } else if (a->opcode() == IrOpcode::kCheckedTaggedSignedToInt32 &&
               b->opcode() == IrOpcode::kCheckedTaggedToArrayIndex) {
      // CheckedTaggedSignedToInt32(node) implies
      // CheckedTaggedToArrayIndex(node)
      if (machine->Is64()) {
        // On 64 bit architectures, ArrayIndex is 64 bit.
        subsumption =
            Subsumption::WithConversion(machine->ChangeInt32ToInt64());
      }
    } else if (a->opcode() == IrOpcode::kCheckedTaggedToInt32 &&
               b->opcode() == IrOpcode::kCheckedTaggedToArrayIndex) {
      // CheckedTaggedToInt32(node) implies CheckedTaggedToArrayIndex(node)
      if (machine->Is64()) {
        // On 64 bit architectures, ArrayIndex is 64 bit.
        subsumption =
            Subsumption::WithConversion(machine->ChangeInt32ToInt64());
      }
    } else if (a->opcode() == IrOpcode::kCheckReceiver &&
               b->opcode() == IrOpcode::kCheckReceiverOrNullOrUndefined) {
      // CheckReceiver(node) implies CheckReceiverOrNullOrUndefined(node)
    } else if (a->opcode() != b->opcode()) {
      return Subsumption::None();
    } else {
      switch (a->opcode()) {
        case IrOpcode::kCheckBounds:
        case IrOpcode::kCheckSmi:
        case IrOpcode::kCheckString:
        case IrOpcode::kCheckStringOrStringWrapper:
        case IrOpcode::kCheckNumber:
        case IrOpcode::kCheckNumberFitsInt32:
        case IrOpcode::kCheckBigInt:
        case IrOpcode::kCheckedBigIntToBigInt64:
          break;
        case IrOpcode::kCheckedInt32ToTaggedSigned:
        case IrOpcode::kCheckedInt64ToInt32:
        case IrOpcode::kCheckedInt64ToTaggedSigned:
        case IrOpcode::kCheckedTaggedSignedToInt32:
        case IrOpcode::kCheckedTaggedToTaggedPointer:
        case IrOpcode::kCheckedTaggedToTaggedSigned:
        case IrOpcode::kCheckedTaggedToArrayIndex:
        case IrOpcode::kCheckedUint32Bounds:
        case IrOpcode::kCheckedUint32ToInt32:
        case IrOpcode::kCheckedUint32ToTaggedSigned:
        case IrOpcode::kCheckedUint64Bounds:
        case IrOpcode::kCheckedUint64ToInt32:
        case IrOpcode::kCheckedUint64ToTaggedSigned:
          break;
        case IrOpcode::kCheckedFloat64ToInt32:
        case IrOpcode::kCheckedFloat64ToAdditiveSafeInteger:
        case IrOpcode::kCheckedFloat64ToInt64:
        case IrOpcode::kCheckedTaggedToInt32:
        case IrOpcode::kCheckedTaggedToAdditiveSafeInteger:
        case IrOpcode::kCheckedTaggedToInt64: {
          const CheckMinusZeroParameters& ap =
              CheckMinusZeroParametersOf(a->op());
          const CheckMinusZeroParameters& bp =
              CheckMinusZeroParametersOf(b->op());
          if (ap.mode() != bp.mode()) {
            return Subsumption::None();
          }
          break;
        }
        case IrOpcode::kCheckedTaggedToFloat64:
        case IrOpcode::kCheckedTruncateTaggedToWord32: {
          CheckTaggedInputParameters const& ap =
              CheckTaggedInputParametersOf(a->op());
          CheckTaggedInputParameters const& bp =
              CheckTaggedInputParametersOf(b->op());
          // {a} subsumes {b} if the modes are either the same, or {a} checks
          // for Number, in which case {b} will be subsumed no matter what.
          if (ap.mode() != bp.mode() &&
              ap.mode() != CheckTaggedInputMode::kNumber) {
            return Subsumption::None();
          }
          break;
        }
        default:
          DCHECK(!IsCheckedWithFeedback(a->op()));
          return Subsumption::None();
      }
    }
  }
  for (int i = a->op()->ValueInputCount(); --i >= 0;) {
    if (a->InputAt(i) != b->InputAt(i)) return Subsumption::None();
  }
  return subsumption;
}
 
bool TypeSubsumes(Node* node, Node* replacement) {
  if (!NodeProperties::IsTyped(node) || !NodeProperties::IsTyped(replacement)) {
    // If either node is untyped, we are running during an untyped optimization
    // phase, and replacement is OK.
    return true;
  }
  Type node_type = NodeProperties::GetType(node);
  Type replacement_type = NodeProperties::GetType(replacement);
  return replacement_type.Is(node_type);
}
 
}  // namespace
 
Node* RedundancyElimination::EffectPathChecks::LookupCheck(
    Node* node, JSGraph* jsgraph) const {
  for (Check const* check = head_; check != nullptr; check = check->next) {
    Subsumption subsumption =
        CheckSubsumes(check->node, node, jsgraph->machine());
    if (!subsumption.IsNone() && TypeSubsumes(node, check->node)) {
      DCHECK(!check->node->IsDead());
      Node* result = check->node;
      if (subsumption.IsWithConversion()) {
        result = jsgraph->graph()->NewNode(subsumption.conversion_operator(),
                                           result);
      }
      return result;
    }
  }
  return nullptr;
}
 
Node* RedundancyElimination::EffectPathChecks::LookupBoundsCheckFor(
    Node* node) const {
  for (Check const* check = head_; check != nullptr; check = check->next) {
    if (check->node->opcode() == IrOpcode::kCheckBounds &&
        check->node->InputAt(0) == node && TypeSubsumes(node, check->node) &&
        !(CheckBoundsParametersOf(check->node->op()).flags() &
          CheckBoundsFlag::kConvertStringAndMinusZero)) {
      return check->node;
    }
  }
  return nullptr;
}
 
RedundancyElimination::EffectPathChecks const*
RedundancyElimination::PathChecksForEffectNodes::Get(Node* node) const {
  size_t const id = node->id();
  if (id < info_for_node_.size()) return info_for_node_[id];
  return nullptr;
}
 
void RedundancyElimination::PathChecksForEffectNodes::Set(
    Node* node, EffectPathChecks const* checks) {
  size_t const id = node->id();
  if (id >= info_for_node_.size()) info_for_node_.resize(id + 1, nullptr);
  info_for_node_[id] = checks;
}
 
Reduction RedundancyElimination::ReduceCheckNode(Node* node) {
  Node* const effect = NodeProperties::GetEffectInput(node);
  EffectPathChecks const* checks = node_checks_.Get(effect);
  // If we do not know anything about the predecessor, do not propagate just yet
  // because we will have to recompute anyway once we compute the predecessor.
  if (checks == nullptr) return NoChange();
  // See if we have another check that dominates us.
  if (Node* check = checks->LookupCheck(node, jsgraph_)) {
    ReplaceWithValue(node, check);
    return Replace(check);
  }
 
  // Learn from this check.
  return UpdateChecks(node, checks->AddCheck(zone(), node));
}
 
Reduction RedundancyElimination::ReduceEffectPhi(Node* node) {
  Node* const control = NodeProperties::GetControlInput(node);
  if (control->opcode() == IrOpcode::kLoop) {
    // Here we rely on having only reducible loops:
    // The loop entry edge always dominates the header, so we can just use
    // the information from the loop entry edge.
    return TakeChecksFromFirstEffect(node);
  }
  DCHECK_EQ(IrOpcode::kMerge, control->opcode());
 
  // Shortcut for the case when we do not know anything about some input.
  int const input_count = node->op()->EffectInputCount();
  for (int i = 0; i < input_count; ++i) {
    Node* const effect = NodeProperties::GetEffectInput(node, i);
    if (node_checks_.Get(effect) == nullptr) return NoChange();
  }
 
  // Make a copy of the first input's checks and merge with the checks
  // from other inputs.
  EffectPathChecks* checks = EffectPathChecks::Copy(
      zone(), node_checks_.Get(NodeProperties::GetEffectInput(node, 0)));
  for (int i = 1; i < input_count; ++i) {
    Node* const input = NodeProperties::GetEffectInput(node, i);
    checks->Merge(node_checks_.Get(input));
  }
  return UpdateChecks(node, checks);
}
 
Reduction RedundancyElimination::ReduceSpeculativeNumberComparison(Node* node) {
  NumberOperationHint const hint = NumberOperationHintOf(node->op());
  Node* const first = NodeProperties::GetValueInput(node, 0);
  Type const first_type = NodeProperties::GetType(first);
  Node* const second = NodeProperties::GetValueInput(node, 1);
  Type const second_type = NodeProperties::GetType(second);
  Node* const effect = NodeProperties::GetEffectInput(node);
  EffectPathChecks const* checks = node_checks_.Get(effect);
 
  // If we do not know anything about the predecessor, do not propagate just yet
  // because we will have to recompute anyway once we compute the predecessor.
  if (checks == nullptr) return NoChange();
 
  // Avoid the potentially expensive lookups below if the {node}
  // has seen non-Smi inputs in the past, which is a clear signal
  // that the comparison is probably not performed on a value that
  // already passed an array bounds check.
  if (hint == NumberOperationHint::kSignedSmall) {
    // Don't bother trying to find a CheckBounds for the {first} input
    // if it's type is already in UnsignedSmall range, since the bounds
    // check is only going to narrow that range further, but the result
    // is not going to make the representation selection any better.
    if (!first_type.Is(Type::UnsignedSmall())) {
      if (Node* check = checks->LookupBoundsCheckFor(first)) {
        if (!first_type.Is(NodeProperties::GetType(check))) {
          // Replace the {first} input with the {check}. This is safe,
          // despite the fact that {check} can truncate -0 to 0, because
          // the regular Number comparisons in JavaScript also identify
          // 0 and -0 (unlike special comparisons as Object.is).
          NodeProperties::ReplaceValueInput(node, check, 0);
          return Changed(node).FollowedBy(
              ReduceSpeculativeNumberComparison(node));
        }
      }
    }
 
    // Don't bother trying to find a CheckBounds for the {second} input
    // if it's type is already in UnsignedSmall range, since the bounds
    // check is only going to narrow that range further, but the result
    // is not going to make the representation selection any better.
    if (!second_type.Is(Type::UnsignedSmall())) {
      if (Node* check = checks->LookupBoundsCheckFor(second)) {
        if (!second_type.Is(NodeProperties::GetType(check))) {
          // Replace the {second} input with the {check}. This is safe,
          // despite the fact that {check} can truncate -0 to 0, because
          // the regular Number comparisons in JavaScript also identify
          // 0 and -0 (unlike special comparisons as Object.is).
          NodeProperties::ReplaceValueInput(node, check, 1);
          return Changed(node).FollowedBy(
              ReduceSpeculativeNumberComparison(node));
        }
      }
    }
  }
 
  return UpdateChecks(node, checks);
}
 
Reduction RedundancyElimination::ReduceSpeculativeNumberOperation(Node* node) {
  DCHECK(node->opcode() == IrOpcode::kSpeculativeNumberAdd ||
         node->opcode() == IrOpcode::kSpeculativeNumberSubtract ||
         node->opcode() == IrOpcode::kSpeculativeAdditiveSafeIntegerAdd ||
         node->opcode() == IrOpcode::kSpeculativeAdditiveSafeIntegerSubtract ||
         node->opcode() == IrOpcode::kSpeculativeSmallIntegerAdd ||
         node->opcode() == IrOpcode::kSpeculativeSmallIntegerSubtract ||
         node->opcode() == IrOpcode::kSpeculativeToNumber);
  DCHECK_EQ(1, node->op()->EffectInputCount());
  DCHECK_EQ(1, node->op()->EffectOutputCount());
 
  Node* const first = NodeProperties::GetValueInput(node, 0);
  Node* const effect = NodeProperties::GetEffectInput(node);
  EffectPathChecks const* checks = node_checks_.Get(effect);
  // If we do not know anything about the predecessor, do not propagate just yet
  // because we will have to recompute anyway once we compute the predecessor.
  if (checks == nullptr) return NoChange();
 
  // Check if there's a CheckBounds operation on {first}
  // in the graph already, which we might be able to
  // reuse here to improve the representation selection
  // for the {node} later on.
  if (Node* check = checks->LookupBoundsCheckFor(first)) {
    // Only use the bounds {check} if its type is better
    // than the type of the {first} node, otherwise we
    // would end up replacing NumberConstant inputs with
    // CheckBounds operations, which is kind of pointless.
    if (!NodeProperties::GetType(first).Is(NodeProperties::GetType(check))) {
      NodeProperties::ReplaceValueInput(node, check, 0);
    }
  }
 
  return UpdateChecks(node, checks);
}
 
Reduction RedundancyElimination::ReduceStart(Node* node) {
  return UpdateChecks(node, EffectPathChecks::Empty(zone()));
}
 
Reduction RedundancyElimination::ReduceOtherNode(Node* node) {
  if (node->op()->EffectInputCount() == 1) {
    if (node->op()->EffectOutputCount() == 1) {
      return TakeChecksFromFirstEffect(node);
    } else {
      // Effect terminators should be handled specially.
      return NoChange();
    }
  }
  DCHECK_EQ(0, node->op()->EffectInputCount());
  DCHECK_EQ(0, node->op()->EffectOutputCount());
  return NoChange();
}
 
Reduction RedundancyElimination::TakeChecksFromFirstEffect(Node* node) {
  DCHECK_EQ(1, node->op()->EffectOutputCount());
  Node* const effect = NodeProperties::GetEffectInput(node);
  EffectPathChecks const* checks = node_checks_.Get(effect);
  // If we do not know anything about the predecessor, do not propagate just yet
  // because we will have to recompute anyway once we compute the predecessor.
  if (checks == nullptr) return NoChange();
  // We just propagate the information from the effect input (ideally,
  // we would only revisit effect uses if there is change).
  return UpdateChecks(node, checks);
}
 
Reduction RedundancyElimination::UpdateChecks(Node* node,
                                              EffectPathChecks const* checks) {
  EffectPathChecks const* original = node_checks_.Get(node);
  // Only signal that the {node} has Changed, if the information about {checks}
  // has changed wrt. the {original}.
  if (checks != original) {
    if (original == nullptr || !checks->Equals(original)) {
      node_checks_.Set(node, checks);
      return Changed(node);
    }
  }
  return NoChange();
}
 
}  // namespace compiler
}  // namespace internal
}  // namespace v8